Automated Improved Human Activity Recognition using Ensemble Modeling

Human Activity Recognition stands as a vital area of research that involves developing systems and algorithms to recognize automatically and classify human activities. This research focuses on the fusion of sensor data and various machine-learning algorithms to recognize and classify human activities. This paper highlights the proficiency of distinct supervised machine learning techniques (DT, RF, SVM, and KNN). The aim of this research is to utilize ensemble learning methodologies, including adaboost, hard and soft voting, and stacking, within the context of human activity recognition. The paper also examines how the results obtained from different machine learning models can be combined, using them as estimators in other ensemble learning approaches like stacking & voting classifiers. The voting classifier, selected after a comprehensive analysis of ensemble learning algorithms, achieved an accuracy of 99.99%. The ensemble model, which integrates four ML models with weight optimization using FOX optimization, gives superior performance, achieving an accuracy of 100%. Furthermore, this paper also addresses the challenges of HAR using smartphones and proposes practical solutions.